Tool Retrieval on ToolBench Standard queries, G1 tier
86.8NDCG@5CoHyDE (r = 3)
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| CoHyDE (r = 3)Expansion Factor (r)=32026.05 | 86.8 | 91 | |
| BGE (trained S1a)Backbone=BGE, Training Stage=S1a2026.05 | 84.2 | 89.7 | |
| HyDE (vanilla LLM + trained BGE S1a)Backbone=BGE (trained S1a), Query Expansion Strategy=HyDE, Generator=Qwen3.5-4B2026.05 | 73.4 | 79.5 | |
| text-embedding-3-largeBackbone=text-embedding-3-large2026.05 | 61.6 | 66.4 | |
| HyDE (vanilla LLM & BGE)Backbone=BGE (vanilla), Query Expansion Strategy=HyDE, Generator=Qwen3.5-4B2026.05 | 58.2 | 63.8 | |
| BGE (vanilla)Backbone=BGE, Training Stage=vanilla2026.05 | 56.5 | 62.4 | |
| Query expansion (LLM + BGE)Backbone=BGE (vanilla), Query Expansion Strategy=Query-side paraphrasing2026.05 | 51.4 | 56.7 | |
| BM252026.05 | 49 | 52.7 |